Research map / Reinforcement learning / Offline RL
Long-Horizon Model-Based Offline Reinforcement Learning Without Explicit Conservatism
Tianwei Ni, Esther Derman, Vineet Jain, Vincent Taboga, Siamak Ravanbakhsh, Pierre-Luc Bacon
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use adaptive uncertainty-based truncation with quantile 1.0 instead of 0.9 (ablation) | NeuBay · D4RL locomotion (12 tasks) | average normalized score: 22.5 -> 80.1 | Improved |
| use truncation quantile 1.0 instead of 0.99 (ablation) | NeuBay · D4RL locomotion (12 tasks) | average normalized score: 45.1 -> 80.1 | Improved |
| use truncation quantile 1.0 instead of 0.999 (ablation) | NeuBay · NeoRL locomotion (9 tasks) | average normalized score: 58.8 -> 64.7 | Improved |
| increase ensemble size from 5 to 100 (ablation) | NeuBay · D4RL locomotion (12 tasks) | average normalized score: 65.9 -> 80.1 | Improved |
| add strong uncertainty penalty (lambda 5.0) to NeuBay (ablation) | NeuBay · D4RL locomotion (12 tasks) | average normalized score: 80.1 -> 57.4 | Hurt |
| add small uncertainty penalty (lambda 0.04) to NeuBay (ablation) | NeuBay · D4RL locomotion (12 tasks) | average normalized score: 80.1 -> 79.8 | No clear effect |
| add uncertainty penalty to NeuBay (dataset-tuned lambda) (ablation) | NeuBay · hopper-random-v2 | normalized score: 24.5 -> 48.2 | Improved |
| add uncertainty penalty to NeuBay (dataset-tuned lambda) (ablation) | NeuBay · hopper-medium-v2 | normalized score: 54.2 -> 105.8 | Improved |
| add uncertainty penalty to NeuBay (ablation) | NeuBay · AntMaze | normalized score | Hurt |
| use history-dependent agent instead of Markov agent (ablation) | NeuBay · D4RL Adroit (3 non-zero tasks) | average normalized score: 36.4 -> 42.2 | Improved |
| use history-dependent agent instead of Markov agent (ablation) | NeuBay · NeoRL locomotion (9 tasks) | average normalized score: 64.7 -> 66.8 (Markov better) | Hurt |
| add layer normalization to the world model (ablation) | NeuBay world model · halfcheetah-medium-expert-v2 | compounding error / predicted state norm | Improved |
Closest papers by meaning
- From Static Policies to Adaptive Priors in Offline Reinforcement Learning (NeurIPS 2026 · similarity 0.92)
- Bayesian Design Principles for Offline-to-Online Reinforcement Learning (ICML 2024 · similarity 0.91)
- OCEAN-MBRL: Offline Conservative Exploration for Model-Based Offline Reinforcement Learning (AAAI 2024 · similarity 0.91)
- Compositional Conservatism: A Transductive Approach in Offline Reinforcement Learning (ICLR 2024 · similarity 0.91)
- Selective Uncertainty Propagation in Offline RL (AAAI 2025 · similarity 0.91)
- Dynamic Uncertainty Estimation for Offline Reinforcement Learning (AAAI 2025 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
